Fixing inconsistent output spaces across studies in fmriprep 1.0.6



Hi, all. We’re trying to combine data from several different scanners using different acquisition sequences with differing voxel sizes. Obviously, this is a problem with the default parameters from fmriprep because the output MNI maps maintain the original voxel dimensions.

If I understand correctly, I can standardize the fmriprep outputs using the --output-grid-reference flag.

I could use help with two questions concerning this:

  1. Which resolution should I choose? I can imagine arguments for choosing the coarsest resolution of all the studies, the finest resolution of all the studies, or a simple 2x2x2 resolution that matches the MNI template defaults.
  2. Is there anything especially wrong with passing the MNI template found within the fmriprep data directory to --output-grid-reference?


Retrospective tagging

I would recommend using the smallest resolution across all scanners to avoid creating files that are too large (with no clear benefits).


After thinking about this a little bit, I think the benefits of a 2x2x2 resolution are twofold for us:

  1. There are a number of atlases that are provided in 2x2x2 resolution, which makes extraction of ROI data very simple without needing to resample either the atlas or the data, and, more importantly,
  2. If this dataset grows to include additional studies from additional scanners, a pre-specified standardized resolution will prevent us from having to reprocess all of the previous studies in the event that a new, smaller-voxelled study comes in.

With that said, I agree with you that the file sizes will be a bit bigger with no clear analytical benefit…


Fair enough. BTW nilearn can handle extracting signals from parcelations in the same space, but different voxel size.